4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/backing-dev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/magic.h>
21 #include <linux/dax.h>
22 #include <linux/buffer_head.h>
23 #include <linux/swap.h>
24 #include <linux/pagevec.h>
25 #include <linux/writeback.h>
26 #include <linux/mpage.h>
27 #include <linux/mount.h>
28 #include <linux/uio.h>
29 #include <linux/namei.h>
30 #include <linux/log2.h>
31 #include <linux/cleancache.h>
32 #include <linux/dax.h>
33 #include <linux/badblocks.h>
34 #include <linux/task_io_accounting_ops.h>
35 #include <linux/falloc.h>
36 #include <linux/uaccess.h>
40 struct block_device bdev
;
41 struct inode vfs_inode
;
44 static const struct address_space_operations def_blk_aops
;
46 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
48 return container_of(inode
, struct bdev_inode
, vfs_inode
);
51 struct block_device
*I_BDEV(struct inode
*inode
)
53 return &BDEV_I(inode
)->bdev
;
55 EXPORT_SYMBOL(I_BDEV
);
57 void __vfs_msg(struct super_block
*sb
, const char *prefix
, const char *fmt
, ...)
65 printk_ratelimited("%sVFS (%s): %pV\n", prefix
, sb
->s_id
, &vaf
);
69 static void bdev_write_inode(struct block_device
*bdev
)
71 struct inode
*inode
= bdev
->bd_inode
;
74 spin_lock(&inode
->i_lock
);
75 while (inode
->i_state
& I_DIRTY
) {
76 spin_unlock(&inode
->i_lock
);
77 ret
= write_inode_now(inode
, true);
79 char name
[BDEVNAME_SIZE
];
80 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
81 "for block device %s (err=%d).\n",
82 bdevname(bdev
, name
), ret
);
84 spin_lock(&inode
->i_lock
);
86 spin_unlock(&inode
->i_lock
);
89 /* Kill _all_ buffers and pagecache , dirty or not.. */
90 void kill_bdev(struct block_device
*bdev
)
92 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
94 if (mapping
->nrpages
== 0 && mapping
->nrexceptional
== 0)
98 truncate_inode_pages(mapping
, 0);
100 EXPORT_SYMBOL(kill_bdev
);
102 /* Invalidate clean unused buffers and pagecache. */
103 void invalidate_bdev(struct block_device
*bdev
)
105 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
107 if (mapping
->nrpages
) {
108 invalidate_bh_lrus();
109 lru_add_drain_all(); /* make sure all lru add caches are flushed */
110 invalidate_mapping_pages(mapping
, 0, -1);
112 /* 99% of the time, we don't need to flush the cleancache on the bdev.
113 * But, for the strange corners, lets be cautious
115 cleancache_invalidate_inode(mapping
);
117 EXPORT_SYMBOL(invalidate_bdev
);
119 int set_blocksize(struct block_device
*bdev
, int size
)
121 /* Size must be a power of two, and between 512 and PAGE_SIZE */
122 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
125 /* Size cannot be smaller than the size supported by the device */
126 if (size
< bdev_logical_block_size(bdev
))
129 /* Don't change the size if it is same as current */
130 if (bdev
->bd_block_size
!= size
) {
132 bdev
->bd_block_size
= size
;
133 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
139 EXPORT_SYMBOL(set_blocksize
);
141 int sb_set_blocksize(struct super_block
*sb
, int size
)
143 if (set_blocksize(sb
->s_bdev
, size
))
145 /* If we get here, we know size is power of two
146 * and it's value is between 512 and PAGE_SIZE */
147 sb
->s_blocksize
= size
;
148 sb
->s_blocksize_bits
= blksize_bits(size
);
149 return sb
->s_blocksize
;
152 EXPORT_SYMBOL(sb_set_blocksize
);
154 int sb_min_blocksize(struct super_block
*sb
, int size
)
156 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
159 return sb_set_blocksize(sb
, size
);
162 EXPORT_SYMBOL(sb_min_blocksize
);
165 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
166 struct buffer_head
*bh
, int create
)
168 bh
->b_bdev
= I_BDEV(inode
);
169 bh
->b_blocknr
= iblock
;
170 set_buffer_mapped(bh
);
174 static struct inode
*bdev_file_inode(struct file
*file
)
176 return file
->f_mapping
->host
;
179 static unsigned int dio_bio_write_op(struct kiocb
*iocb
)
181 unsigned int op
= REQ_OP_WRITE
| REQ_SYNC
| REQ_IDLE
;
183 /* avoid the need for a I/O completion work item */
184 if (iocb
->ki_flags
& IOCB_DSYNC
)
189 #define DIO_INLINE_BIO_VECS 4
191 static void blkdev_bio_end_io_simple(struct bio
*bio
)
193 struct task_struct
*waiter
= bio
->bi_private
;
195 WRITE_ONCE(bio
->bi_private
, NULL
);
196 wake_up_process(waiter
);
200 __blkdev_direct_IO_simple(struct kiocb
*iocb
, struct iov_iter
*iter
,
203 struct file
*file
= iocb
->ki_filp
;
204 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
205 struct bio_vec inline_vecs
[DIO_INLINE_BIO_VECS
], *vecs
, *bvec
;
206 loff_t pos
= iocb
->ki_pos
;
207 bool should_dirty
= false;
213 if ((pos
| iov_iter_alignment(iter
)) &
214 (bdev_logical_block_size(bdev
) - 1))
217 if (nr_pages
<= DIO_INLINE_BIO_VECS
)
220 vecs
= kmalloc(nr_pages
* sizeof(struct bio_vec
), GFP_KERNEL
);
225 bio_init(&bio
, vecs
, nr_pages
);
227 bio
.bi_iter
.bi_sector
= pos
>> 9;
228 bio
.bi_private
= current
;
229 bio
.bi_end_io
= blkdev_bio_end_io_simple
;
231 ret
= bio_iov_iter_get_pages(&bio
, iter
);
234 ret
= bio
.bi_iter
.bi_size
;
236 if (iov_iter_rw(iter
) == READ
) {
237 bio
.bi_opf
= REQ_OP_READ
;
238 if (iter_is_iovec(iter
))
241 bio
.bi_opf
= dio_bio_write_op(iocb
);
242 task_io_account_write(ret
);
245 qc
= submit_bio(&bio
);
247 set_current_state(TASK_UNINTERRUPTIBLE
);
248 if (!READ_ONCE(bio
.bi_private
))
250 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
251 !blk_mq_poll(bdev_get_queue(bdev
), qc
))
254 __set_current_state(TASK_RUNNING
);
256 bio_for_each_segment_all(bvec
, &bio
, i
) {
257 if (should_dirty
&& !PageCompound(bvec
->bv_page
))
258 set_page_dirty_lock(bvec
->bv_page
);
259 put_page(bvec
->bv_page
);
262 if (vecs
!= inline_vecs
)
265 if (unlikely(bio
.bi_error
))
273 struct task_struct
*waiter
;
278 bool should_dirty
: 1;
283 static struct bio_set
*blkdev_dio_pool __read_mostly
;
285 static void blkdev_bio_end_io(struct bio
*bio
)
287 struct blkdev_dio
*dio
= bio
->bi_private
;
288 bool should_dirty
= dio
->should_dirty
;
290 if (dio
->multi_bio
&& !atomic_dec_and_test(&dio
->ref
)) {
291 if (bio
->bi_error
&& !dio
->bio
.bi_error
)
292 dio
->bio
.bi_error
= bio
->bi_error
;
295 struct kiocb
*iocb
= dio
->iocb
;
296 ssize_t ret
= dio
->bio
.bi_error
;
303 dio
->iocb
->ki_complete(iocb
, ret
, 0);
306 struct task_struct
*waiter
= dio
->waiter
;
308 WRITE_ONCE(dio
->waiter
, NULL
);
309 wake_up_process(waiter
);
314 bio_check_pages_dirty(bio
);
316 struct bio_vec
*bvec
;
319 bio_for_each_segment_all(bvec
, bio
, i
)
320 put_page(bvec
->bv_page
);
326 __blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
, int nr_pages
)
328 struct file
*file
= iocb
->ki_filp
;
329 struct inode
*inode
= bdev_file_inode(file
);
330 struct block_device
*bdev
= I_BDEV(inode
);
331 struct blk_plug plug
;
332 struct blkdev_dio
*dio
;
334 bool is_read
= (iov_iter_rw(iter
) == READ
), is_sync
;
335 loff_t pos
= iocb
->ki_pos
;
336 blk_qc_t qc
= BLK_QC_T_NONE
;
339 if ((pos
| iov_iter_alignment(iter
)) &
340 (bdev_logical_block_size(bdev
) - 1))
343 bio
= bio_alloc_bioset(GFP_KERNEL
, nr_pages
, blkdev_dio_pool
);
344 bio_get(bio
); /* extra ref for the completion handler */
346 dio
= container_of(bio
, struct blkdev_dio
, bio
);
347 dio
->is_sync
= is_sync
= is_sync_kiocb(iocb
);
349 dio
->waiter
= current
;
354 dio
->multi_bio
= false;
355 dio
->should_dirty
= is_read
&& (iter
->type
== ITER_IOVEC
);
357 blk_start_plug(&plug
);
360 bio
->bi_iter
.bi_sector
= pos
>> 9;
361 bio
->bi_private
= dio
;
362 bio
->bi_end_io
= blkdev_bio_end_io
;
364 ret
= bio_iov_iter_get_pages(bio
, iter
);
372 bio
->bi_opf
= REQ_OP_READ
;
373 if (dio
->should_dirty
)
374 bio_set_pages_dirty(bio
);
376 bio
->bi_opf
= dio_bio_write_op(iocb
);
377 task_io_account_write(bio
->bi_iter
.bi_size
);
380 dio
->size
+= bio
->bi_iter
.bi_size
;
381 pos
+= bio
->bi_iter
.bi_size
;
383 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
);
385 qc
= submit_bio(bio
);
389 if (!dio
->multi_bio
) {
390 dio
->multi_bio
= true;
391 atomic_set(&dio
->ref
, 2);
393 atomic_inc(&dio
->ref
);
397 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
399 blk_finish_plug(&plug
);
405 set_current_state(TASK_UNINTERRUPTIBLE
);
406 if (!READ_ONCE(dio
->waiter
))
409 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
410 !blk_mq_poll(bdev_get_queue(bdev
), qc
))
413 __set_current_state(TASK_RUNNING
);
415 ret
= dio
->bio
.bi_error
;
424 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
428 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
+ 1);
431 if (is_sync_kiocb(iocb
) && nr_pages
<= BIO_MAX_PAGES
)
432 return __blkdev_direct_IO_simple(iocb
, iter
, nr_pages
);
434 return __blkdev_direct_IO(iocb
, iter
, min(nr_pages
, BIO_MAX_PAGES
));
437 static __init
int blkdev_init(void)
439 blkdev_dio_pool
= bioset_create(4, offsetof(struct blkdev_dio
, bio
));
440 if (!blkdev_dio_pool
)
444 module_init(blkdev_init
);
446 int __sync_blockdev(struct block_device
*bdev
, int wait
)
451 return filemap_flush(bdev
->bd_inode
->i_mapping
);
452 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
456 * Write out and wait upon all the dirty data associated with a block
457 * device via its mapping. Does not take the superblock lock.
459 int sync_blockdev(struct block_device
*bdev
)
461 return __sync_blockdev(bdev
, 1);
463 EXPORT_SYMBOL(sync_blockdev
);
466 * Write out and wait upon all dirty data associated with this
467 * device. Filesystem data as well as the underlying block
468 * device. Takes the superblock lock.
470 int fsync_bdev(struct block_device
*bdev
)
472 struct super_block
*sb
= get_super(bdev
);
474 int res
= sync_filesystem(sb
);
478 return sync_blockdev(bdev
);
480 EXPORT_SYMBOL(fsync_bdev
);
483 * freeze_bdev -- lock a filesystem and force it into a consistent state
484 * @bdev: blockdevice to lock
486 * If a superblock is found on this device, we take the s_umount semaphore
487 * on it to make sure nobody unmounts until the snapshot creation is done.
488 * The reference counter (bd_fsfreeze_count) guarantees that only the last
489 * unfreeze process can unfreeze the frozen filesystem actually when multiple
490 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
491 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
494 struct super_block
*freeze_bdev(struct block_device
*bdev
)
496 struct super_block
*sb
;
499 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
500 if (++bdev
->bd_fsfreeze_count
> 1) {
502 * We don't even need to grab a reference - the first call
503 * to freeze_bdev grab an active reference and only the last
504 * thaw_bdev drops it.
506 sb
= get_super(bdev
);
509 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
513 sb
= get_active_super(bdev
);
516 if (sb
->s_op
->freeze_super
)
517 error
= sb
->s_op
->freeze_super(sb
);
519 error
= freeze_super(sb
);
521 deactivate_super(sb
);
522 bdev
->bd_fsfreeze_count
--;
523 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
524 return ERR_PTR(error
);
526 deactivate_super(sb
);
529 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
530 return sb
; /* thaw_bdev releases s->s_umount */
532 EXPORT_SYMBOL(freeze_bdev
);
535 * thaw_bdev -- unlock filesystem
536 * @bdev: blockdevice to unlock
537 * @sb: associated superblock
539 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
541 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
545 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
546 if (!bdev
->bd_fsfreeze_count
)
550 if (--bdev
->bd_fsfreeze_count
> 0)
556 if (sb
->s_op
->thaw_super
)
557 error
= sb
->s_op
->thaw_super(sb
);
559 error
= thaw_super(sb
);
561 bdev
->bd_fsfreeze_count
++;
563 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
566 EXPORT_SYMBOL(thaw_bdev
);
568 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
570 return block_write_full_page(page
, blkdev_get_block
, wbc
);
573 static int blkdev_readpage(struct file
* file
, struct page
* page
)
575 return block_read_full_page(page
, blkdev_get_block
);
578 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
579 struct list_head
*pages
, unsigned nr_pages
)
581 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
584 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
585 loff_t pos
, unsigned len
, unsigned flags
,
586 struct page
**pagep
, void **fsdata
)
588 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
592 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
593 loff_t pos
, unsigned len
, unsigned copied
,
594 struct page
*page
, void *fsdata
)
597 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
607 * for a block special file file_inode(file)->i_size is zero
608 * so we compute the size by hand (just as in block_read/write above)
610 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
612 struct inode
*bd_inode
= bdev_file_inode(file
);
615 inode_lock(bd_inode
);
616 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
617 inode_unlock(bd_inode
);
621 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
623 struct inode
*bd_inode
= bdev_file_inode(filp
);
624 struct block_device
*bdev
= I_BDEV(bd_inode
);
627 error
= filemap_write_and_wait_range(filp
->f_mapping
, start
, end
);
632 * There is no need to serialise calls to blkdev_issue_flush with
633 * i_mutex and doing so causes performance issues with concurrent
634 * O_SYNC writers to a block device.
636 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
637 if (error
== -EOPNOTSUPP
)
642 EXPORT_SYMBOL(blkdev_fsync
);
645 * bdev_read_page() - Start reading a page from a block device
646 * @bdev: The device to read the page from
647 * @sector: The offset on the device to read the page to (need not be aligned)
648 * @page: The page to read
650 * On entry, the page should be locked. It will be unlocked when the page
651 * has been read. If the block driver implements rw_page synchronously,
652 * that will be true on exit from this function, but it need not be.
654 * Errors returned by this function are usually "soft", eg out of memory, or
655 * queue full; callers should try a different route to read this page rather
656 * than propagate an error back up the stack.
658 * Return: negative errno if an error occurs, 0 if submission was successful.
660 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
663 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
664 int result
= -EOPNOTSUPP
;
666 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
669 result
= blk_queue_enter(bdev
->bd_queue
, false);
672 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, false);
673 blk_queue_exit(bdev
->bd_queue
);
676 EXPORT_SYMBOL_GPL(bdev_read_page
);
679 * bdev_write_page() - Start writing a page to a block device
680 * @bdev: The device to write the page to
681 * @sector: The offset on the device to write the page to (need not be aligned)
682 * @page: The page to write
683 * @wbc: The writeback_control for the write
685 * On entry, the page should be locked and not currently under writeback.
686 * On exit, if the write started successfully, the page will be unlocked and
687 * under writeback. If the write failed already (eg the driver failed to
688 * queue the page to the device), the page will still be locked. If the
689 * caller is a ->writepage implementation, it will need to unlock the page.
691 * Errors returned by this function are usually "soft", eg out of memory, or
692 * queue full; callers should try a different route to write this page rather
693 * than propagate an error back up the stack.
695 * Return: negative errno if an error occurs, 0 if submission was successful.
697 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
698 struct page
*page
, struct writeback_control
*wbc
)
701 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
703 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
705 result
= blk_queue_enter(bdev
->bd_queue
, false);
709 set_page_writeback(page
);
710 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, true);
712 end_page_writeback(page
);
715 blk_queue_exit(bdev
->bd_queue
);
718 EXPORT_SYMBOL_GPL(bdev_write_page
);
724 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
725 static struct kmem_cache
* bdev_cachep __read_mostly
;
727 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
729 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
732 return &ei
->vfs_inode
;
735 static void bdev_i_callback(struct rcu_head
*head
)
737 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
738 struct bdev_inode
*bdi
= BDEV_I(inode
);
740 kmem_cache_free(bdev_cachep
, bdi
);
743 static void bdev_destroy_inode(struct inode
*inode
)
745 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
748 static void init_once(void *foo
)
750 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
751 struct block_device
*bdev
= &ei
->bdev
;
753 memset(bdev
, 0, sizeof(*bdev
));
754 mutex_init(&bdev
->bd_mutex
);
755 INIT_LIST_HEAD(&bdev
->bd_list
);
757 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
759 bdev
->bd_bdi
= &noop_backing_dev_info
;
760 inode_init_once(&ei
->vfs_inode
);
761 /* Initialize mutex for freeze. */
762 mutex_init(&bdev
->bd_fsfreeze_mutex
);
765 static void bdev_evict_inode(struct inode
*inode
)
767 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
768 truncate_inode_pages_final(&inode
->i_data
);
769 invalidate_inode_buffers(inode
); /* is it needed here? */
771 spin_lock(&bdev_lock
);
772 list_del_init(&bdev
->bd_list
);
773 spin_unlock(&bdev_lock
);
774 /* Detach inode from wb early as bdi_put() may free bdi->wb */
775 inode_detach_wb(inode
);
776 if (bdev
->bd_bdi
!= &noop_backing_dev_info
) {
777 bdi_put(bdev
->bd_bdi
);
778 bdev
->bd_bdi
= &noop_backing_dev_info
;
782 static const struct super_operations bdev_sops
= {
783 .statfs
= simple_statfs
,
784 .alloc_inode
= bdev_alloc_inode
,
785 .destroy_inode
= bdev_destroy_inode
,
786 .drop_inode
= generic_delete_inode
,
787 .evict_inode
= bdev_evict_inode
,
790 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
791 int flags
, const char *dev_name
, void *data
)
794 dent
= mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
796 dent
->d_sb
->s_iflags
|= SB_I_CGROUPWB
;
800 static struct file_system_type bd_type
= {
803 .kill_sb
= kill_anon_super
,
806 struct super_block
*blockdev_superblock __read_mostly
;
807 EXPORT_SYMBOL_GPL(blockdev_superblock
);
809 void __init
bdev_cache_init(void)
812 static struct vfsmount
*bd_mnt
;
814 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
815 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
816 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
|SLAB_PANIC
),
818 err
= register_filesystem(&bd_type
);
820 panic("Cannot register bdev pseudo-fs");
821 bd_mnt
= kern_mount(&bd_type
);
823 panic("Cannot create bdev pseudo-fs");
824 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
828 * Most likely _very_ bad one - but then it's hardly critical for small
829 * /dev and can be fixed when somebody will need really large one.
830 * Keep in mind that it will be fed through icache hash function too.
832 static inline unsigned long hash(dev_t dev
)
834 return MAJOR(dev
)+MINOR(dev
);
837 static int bdev_test(struct inode
*inode
, void *data
)
839 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
842 static int bdev_set(struct inode
*inode
, void *data
)
844 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
848 static LIST_HEAD(all_bdevs
);
851 * If there is a bdev inode for this device, unhash it so that it gets evicted
852 * as soon as last inode reference is dropped.
854 void bdev_unhash_inode(dev_t dev
)
858 inode
= ilookup5(blockdev_superblock
, hash(dev
), bdev_test
, &dev
);
860 remove_inode_hash(inode
);
865 struct block_device
*bdget(dev_t dev
)
867 struct block_device
*bdev
;
870 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
871 bdev_test
, bdev_set
, &dev
);
876 bdev
= &BDEV_I(inode
)->bdev
;
878 if (inode
->i_state
& I_NEW
) {
879 bdev
->bd_contains
= NULL
;
880 bdev
->bd_super
= NULL
;
881 bdev
->bd_inode
= inode
;
882 bdev
->bd_block_size
= i_blocksize(inode
);
883 bdev
->bd_part_count
= 0;
884 bdev
->bd_invalidated
= 0;
885 inode
->i_mode
= S_IFBLK
;
887 inode
->i_bdev
= bdev
;
888 inode
->i_data
.a_ops
= &def_blk_aops
;
889 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
890 spin_lock(&bdev_lock
);
891 list_add(&bdev
->bd_list
, &all_bdevs
);
892 spin_unlock(&bdev_lock
);
893 unlock_new_inode(inode
);
898 EXPORT_SYMBOL(bdget
);
901 * bdgrab -- Grab a reference to an already referenced block device
902 * @bdev: Block device to grab a reference to.
904 struct block_device
*bdgrab(struct block_device
*bdev
)
906 ihold(bdev
->bd_inode
);
909 EXPORT_SYMBOL(bdgrab
);
911 long nr_blockdev_pages(void)
913 struct block_device
*bdev
;
915 spin_lock(&bdev_lock
);
916 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
917 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
919 spin_unlock(&bdev_lock
);
923 void bdput(struct block_device
*bdev
)
925 iput(bdev
->bd_inode
);
928 EXPORT_SYMBOL(bdput
);
930 static struct block_device
*bd_acquire(struct inode
*inode
)
932 struct block_device
*bdev
;
934 spin_lock(&bdev_lock
);
935 bdev
= inode
->i_bdev
;
936 if (bdev
&& !inode_unhashed(bdev
->bd_inode
)) {
938 spin_unlock(&bdev_lock
);
941 spin_unlock(&bdev_lock
);
944 * i_bdev references block device inode that was already shut down
945 * (corresponding device got removed). Remove the reference and look
946 * up block device inode again just in case new device got
947 * reestablished under the same device number.
952 bdev
= bdget(inode
->i_rdev
);
954 spin_lock(&bdev_lock
);
955 if (!inode
->i_bdev
) {
957 * We take an additional reference to bd_inode,
958 * and it's released in clear_inode() of inode.
959 * So, we can access it via ->i_mapping always
963 inode
->i_bdev
= bdev
;
964 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
966 spin_unlock(&bdev_lock
);
971 /* Call when you free inode */
973 void bd_forget(struct inode
*inode
)
975 struct block_device
*bdev
= NULL
;
977 spin_lock(&bdev_lock
);
978 if (!sb_is_blkdev_sb(inode
->i_sb
))
979 bdev
= inode
->i_bdev
;
980 inode
->i_bdev
= NULL
;
981 inode
->i_mapping
= &inode
->i_data
;
982 spin_unlock(&bdev_lock
);
989 * bd_may_claim - test whether a block device can be claimed
990 * @bdev: block device of interest
991 * @whole: whole block device containing @bdev, may equal @bdev
992 * @holder: holder trying to claim @bdev
994 * Test whether @bdev can be claimed by @holder.
997 * spin_lock(&bdev_lock).
1000 * %true if @bdev can be claimed, %false otherwise.
1002 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
1005 if (bdev
->bd_holder
== holder
)
1006 return true; /* already a holder */
1007 else if (bdev
->bd_holder
!= NULL
)
1008 return false; /* held by someone else */
1009 else if (whole
== bdev
)
1010 return true; /* is a whole device which isn't held */
1012 else if (whole
->bd_holder
== bd_may_claim
)
1013 return true; /* is a partition of a device that is being partitioned */
1014 else if (whole
->bd_holder
!= NULL
)
1015 return false; /* is a partition of a held device */
1017 return true; /* is a partition of an un-held device */
1021 * bd_prepare_to_claim - prepare to claim a block device
1022 * @bdev: block device of interest
1023 * @whole: the whole device containing @bdev, may equal @bdev
1024 * @holder: holder trying to claim @bdev
1026 * Prepare to claim @bdev. This function fails if @bdev is already
1027 * claimed by another holder and waits if another claiming is in
1028 * progress. This function doesn't actually claim. On successful
1029 * return, the caller has ownership of bd_claiming and bd_holder[s].
1032 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1033 * it multiple times.
1036 * 0 if @bdev can be claimed, -EBUSY otherwise.
1038 static int bd_prepare_to_claim(struct block_device
*bdev
,
1039 struct block_device
*whole
, void *holder
)
1042 /* if someone else claimed, fail */
1043 if (!bd_may_claim(bdev
, whole
, holder
))
1046 /* if claiming is already in progress, wait for it to finish */
1047 if (whole
->bd_claiming
) {
1048 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
1051 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
1052 spin_unlock(&bdev_lock
);
1054 finish_wait(wq
, &wait
);
1055 spin_lock(&bdev_lock
);
1064 * bd_start_claiming - start claiming a block device
1065 * @bdev: block device of interest
1066 * @holder: holder trying to claim @bdev
1068 * @bdev is about to be opened exclusively. Check @bdev can be opened
1069 * exclusively and mark that an exclusive open is in progress. Each
1070 * successful call to this function must be matched with a call to
1071 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1074 * This function is used to gain exclusive access to the block device
1075 * without actually causing other exclusive open attempts to fail. It
1076 * should be used when the open sequence itself requires exclusive
1077 * access but may subsequently fail.
1083 * Pointer to the block device containing @bdev on success, ERR_PTR()
1086 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
1089 struct gendisk
*disk
;
1090 struct block_device
*whole
;
1096 * @bdev might not have been initialized properly yet, look up
1097 * and grab the outer block device the hard way.
1099 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1101 return ERR_PTR(-ENXIO
);
1104 * Normally, @bdev should equal what's returned from bdget_disk()
1105 * if partno is 0; however, some drivers (floppy) use multiple
1106 * bdev's for the same physical device and @bdev may be one of the
1107 * aliases. Keep @bdev if partno is 0. This means claimer
1108 * tracking is broken for those devices but it has always been that
1112 whole
= bdget_disk(disk
, 0);
1114 whole
= bdgrab(bdev
);
1116 module_put(disk
->fops
->owner
);
1119 return ERR_PTR(-ENOMEM
);
1121 /* prepare to claim, if successful, mark claiming in progress */
1122 spin_lock(&bdev_lock
);
1124 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
1126 whole
->bd_claiming
= holder
;
1127 spin_unlock(&bdev_lock
);
1130 spin_unlock(&bdev_lock
);
1132 return ERR_PTR(err
);
1137 struct bd_holder_disk
{
1138 struct list_head list
;
1139 struct gendisk
*disk
;
1143 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
1144 struct gendisk
*disk
)
1146 struct bd_holder_disk
*holder
;
1148 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
1149 if (holder
->disk
== disk
)
1154 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
1156 return sysfs_create_link(from
, to
, kobject_name(to
));
1159 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
1161 sysfs_remove_link(from
, kobject_name(to
));
1165 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1166 * @bdev: the claimed slave bdev
1167 * @disk: the holding disk
1169 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1171 * This functions creates the following sysfs symlinks.
1173 * - from "slaves" directory of the holder @disk to the claimed @bdev
1174 * - from "holders" directory of the @bdev to the holder @disk
1176 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1177 * passed to bd_link_disk_holder(), then:
1179 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1180 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1182 * The caller must have claimed @bdev before calling this function and
1183 * ensure that both @bdev and @disk are valid during the creation and
1184 * lifetime of these symlinks.
1190 * 0 on success, -errno on failure.
1192 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1194 struct bd_holder_disk
*holder
;
1197 mutex_lock(&bdev
->bd_mutex
);
1199 WARN_ON_ONCE(!bdev
->bd_holder
);
1201 /* FIXME: remove the following once add_disk() handles errors */
1202 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
1205 holder
= bd_find_holder_disk(bdev
, disk
);
1211 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
1217 INIT_LIST_HEAD(&holder
->list
);
1218 holder
->disk
= disk
;
1221 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1225 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
1229 * bdev could be deleted beneath us which would implicitly destroy
1230 * the holder directory. Hold on to it.
1232 kobject_get(bdev
->bd_part
->holder_dir
);
1234 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
1238 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1242 mutex_unlock(&bdev
->bd_mutex
);
1245 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1248 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1249 * @bdev: the calimed slave bdev
1250 * @disk: the holding disk
1252 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1257 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1259 struct bd_holder_disk
*holder
;
1261 mutex_lock(&bdev
->bd_mutex
);
1263 holder
= bd_find_holder_disk(bdev
, disk
);
1265 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1266 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1267 del_symlink(bdev
->bd_part
->holder_dir
,
1268 &disk_to_dev(disk
)->kobj
);
1269 kobject_put(bdev
->bd_part
->holder_dir
);
1270 list_del_init(&holder
->list
);
1274 mutex_unlock(&bdev
->bd_mutex
);
1276 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1280 * flush_disk - invalidates all buffer-cache entries on a disk
1282 * @bdev: struct block device to be flushed
1283 * @kill_dirty: flag to guide handling of dirty inodes
1285 * Invalidates all buffer-cache entries on a disk. It should be called
1286 * when a disk has been changed -- either by a media change or online
1289 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1291 if (__invalidate_device(bdev
, kill_dirty
)) {
1292 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1293 "resized disk %s\n",
1294 bdev
->bd_disk
? bdev
->bd_disk
->disk_name
: "");
1299 if (disk_part_scan_enabled(bdev
->bd_disk
))
1300 bdev
->bd_invalidated
= 1;
1304 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1305 * @disk: struct gendisk to check
1306 * @bdev: struct bdev to adjust.
1308 * This routine checks to see if the bdev size does not match the disk size
1309 * and adjusts it if it differs.
1311 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1313 loff_t disk_size
, bdev_size
;
1315 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1316 bdev_size
= i_size_read(bdev
->bd_inode
);
1317 if (disk_size
!= bdev_size
) {
1319 "%s: detected capacity change from %lld to %lld\n",
1320 disk
->disk_name
, bdev_size
, disk_size
);
1321 i_size_write(bdev
->bd_inode
, disk_size
);
1322 flush_disk(bdev
, false);
1325 EXPORT_SYMBOL(check_disk_size_change
);
1328 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1329 * @disk: struct gendisk to be revalidated
1331 * This routine is a wrapper for lower-level driver's revalidate_disk
1332 * call-backs. It is used to do common pre and post operations needed
1333 * for all revalidate_disk operations.
1335 int revalidate_disk(struct gendisk
*disk
)
1337 struct block_device
*bdev
;
1340 if (disk
->fops
->revalidate_disk
)
1341 ret
= disk
->fops
->revalidate_disk(disk
);
1342 bdev
= bdget_disk(disk
, 0);
1346 mutex_lock(&bdev
->bd_mutex
);
1347 check_disk_size_change(disk
, bdev
);
1348 bdev
->bd_invalidated
= 0;
1349 mutex_unlock(&bdev
->bd_mutex
);
1353 EXPORT_SYMBOL(revalidate_disk
);
1356 * This routine checks whether a removable media has been changed,
1357 * and invalidates all buffer-cache-entries in that case. This
1358 * is a relatively slow routine, so we have to try to minimize using
1359 * it. Thus it is called only upon a 'mount' or 'open'. This
1360 * is the best way of combining speed and utility, I think.
1361 * People changing diskettes in the middle of an operation deserve
1364 int check_disk_change(struct block_device
*bdev
)
1366 struct gendisk
*disk
= bdev
->bd_disk
;
1367 const struct block_device_operations
*bdops
= disk
->fops
;
1368 unsigned int events
;
1370 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1371 DISK_EVENT_EJECT_REQUEST
);
1372 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1375 flush_disk(bdev
, true);
1376 if (bdops
->revalidate_disk
)
1377 bdops
->revalidate_disk(bdev
->bd_disk
);
1381 EXPORT_SYMBOL(check_disk_change
);
1383 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1385 unsigned bsize
= bdev_logical_block_size(bdev
);
1387 inode_lock(bdev
->bd_inode
);
1388 i_size_write(bdev
->bd_inode
, size
);
1389 inode_unlock(bdev
->bd_inode
);
1390 while (bsize
< PAGE_SIZE
) {
1395 bdev
->bd_block_size
= bsize
;
1396 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1398 EXPORT_SYMBOL(bd_set_size
);
1400 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1405 * mutex_lock(part->bd_mutex)
1406 * mutex_lock_nested(whole->bd_mutex, 1)
1409 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1411 struct gendisk
*disk
;
1412 struct module
*owner
;
1417 if (mode
& FMODE_READ
)
1419 if (mode
& FMODE_WRITE
)
1422 * hooks: /n/, see "layering violations".
1425 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1435 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1438 owner
= disk
->fops
->owner
;
1440 disk_block_events(disk
);
1441 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1442 if (!bdev
->bd_openers
) {
1443 bdev
->bd_disk
= disk
;
1444 bdev
->bd_queue
= disk
->queue
;
1445 bdev
->bd_contains
= bdev
;
1449 bdev
->bd_part
= disk_get_part(disk
, partno
);
1454 if (disk
->fops
->open
) {
1455 ret
= disk
->fops
->open(bdev
, mode
);
1456 if (ret
== -ERESTARTSYS
) {
1457 /* Lost a race with 'disk' being
1458 * deleted, try again.
1461 disk_put_part(bdev
->bd_part
);
1462 bdev
->bd_part
= NULL
;
1463 bdev
->bd_disk
= NULL
;
1464 bdev
->bd_queue
= NULL
;
1465 mutex_unlock(&bdev
->bd_mutex
);
1466 disk_unblock_events(disk
);
1474 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1477 * If the device is invalidated, rescan partition
1478 * if open succeeded or failed with -ENOMEDIUM.
1479 * The latter is necessary to prevent ghost
1480 * partitions on a removed medium.
1482 if (bdev
->bd_invalidated
) {
1484 rescan_partitions(disk
, bdev
);
1485 else if (ret
== -ENOMEDIUM
)
1486 invalidate_partitions(disk
, bdev
);
1492 struct block_device
*whole
;
1493 whole
= bdget_disk(disk
, 0);
1498 ret
= __blkdev_get(whole
, mode
, 1);
1501 bdev
->bd_contains
= whole
;
1502 bdev
->bd_part
= disk_get_part(disk
, partno
);
1503 if (!(disk
->flags
& GENHD_FL_UP
) ||
1504 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1508 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1511 if (bdev
->bd_bdi
== &noop_backing_dev_info
)
1512 bdev
->bd_bdi
= bdi_get(disk
->queue
->backing_dev_info
);
1514 if (bdev
->bd_contains
== bdev
) {
1516 if (bdev
->bd_disk
->fops
->open
)
1517 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1518 /* the same as first opener case, read comment there */
1519 if (bdev
->bd_invalidated
) {
1521 rescan_partitions(bdev
->bd_disk
, bdev
);
1522 else if (ret
== -ENOMEDIUM
)
1523 invalidate_partitions(bdev
->bd_disk
, bdev
);
1526 goto out_unlock_bdev
;
1528 /* only one opener holds refs to the module and disk */
1534 bdev
->bd_part_count
++;
1535 mutex_unlock(&bdev
->bd_mutex
);
1536 disk_unblock_events(disk
);
1540 disk_put_part(bdev
->bd_part
);
1541 bdev
->bd_disk
= NULL
;
1542 bdev
->bd_part
= NULL
;
1543 bdev
->bd_queue
= NULL
;
1544 if (bdev
!= bdev
->bd_contains
)
1545 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1546 bdev
->bd_contains
= NULL
;
1548 mutex_unlock(&bdev
->bd_mutex
);
1549 disk_unblock_events(disk
);
1559 * blkdev_get - open a block device
1560 * @bdev: block_device to open
1561 * @mode: FMODE_* mask
1562 * @holder: exclusive holder identifier
1564 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1565 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1566 * @holder is invalid. Exclusive opens may nest for the same @holder.
1568 * On success, the reference count of @bdev is unchanged. On failure,
1575 * 0 on success, -errno on failure.
1577 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1579 struct block_device
*whole
= NULL
;
1582 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1584 if ((mode
& FMODE_EXCL
) && holder
) {
1585 whole
= bd_start_claiming(bdev
, holder
);
1586 if (IS_ERR(whole
)) {
1588 return PTR_ERR(whole
);
1592 res
= __blkdev_get(bdev
, mode
, 0);
1595 struct gendisk
*disk
= whole
->bd_disk
;
1597 /* finish claiming */
1598 mutex_lock(&bdev
->bd_mutex
);
1599 spin_lock(&bdev_lock
);
1602 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1604 * Note that for a whole device bd_holders
1605 * will be incremented twice, and bd_holder
1606 * will be set to bd_may_claim before being
1609 whole
->bd_holders
++;
1610 whole
->bd_holder
= bd_may_claim
;
1612 bdev
->bd_holder
= holder
;
1615 /* tell others that we're done */
1616 BUG_ON(whole
->bd_claiming
!= holder
);
1617 whole
->bd_claiming
= NULL
;
1618 wake_up_bit(&whole
->bd_claiming
, 0);
1620 spin_unlock(&bdev_lock
);
1623 * Block event polling for write claims if requested. Any
1624 * write holder makes the write_holder state stick until
1625 * all are released. This is good enough and tracking
1626 * individual writeable reference is too fragile given the
1627 * way @mode is used in blkdev_get/put().
1629 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1630 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1631 bdev
->bd_write_holder
= true;
1632 disk_block_events(disk
);
1635 mutex_unlock(&bdev
->bd_mutex
);
1641 EXPORT_SYMBOL(blkdev_get
);
1644 * blkdev_get_by_path - open a block device by name
1645 * @path: path to the block device to open
1646 * @mode: FMODE_* mask
1647 * @holder: exclusive holder identifier
1649 * Open the blockdevice described by the device file at @path. @mode
1650 * and @holder are identical to blkdev_get().
1652 * On success, the returned block_device has reference count of one.
1658 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1660 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1663 struct block_device
*bdev
;
1666 bdev
= lookup_bdev(path
);
1670 err
= blkdev_get(bdev
, mode
, holder
);
1672 return ERR_PTR(err
);
1674 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1675 blkdev_put(bdev
, mode
);
1676 return ERR_PTR(-EACCES
);
1681 EXPORT_SYMBOL(blkdev_get_by_path
);
1684 * blkdev_get_by_dev - open a block device by device number
1685 * @dev: device number of block device to open
1686 * @mode: FMODE_* mask
1687 * @holder: exclusive holder identifier
1689 * Open the blockdevice described by device number @dev. @mode and
1690 * @holder are identical to blkdev_get().
1692 * Use it ONLY if you really do not have anything better - i.e. when
1693 * you are behind a truly sucky interface and all you are given is a
1694 * device number. _Never_ to be used for internal purposes. If you
1695 * ever need it - reconsider your API.
1697 * On success, the returned block_device has reference count of one.
1703 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1705 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1707 struct block_device
*bdev
;
1712 return ERR_PTR(-ENOMEM
);
1714 err
= blkdev_get(bdev
, mode
, holder
);
1716 return ERR_PTR(err
);
1720 EXPORT_SYMBOL(blkdev_get_by_dev
);
1722 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1724 struct block_device
*bdev
;
1727 * Preserve backwards compatibility and allow large file access
1728 * even if userspace doesn't ask for it explicitly. Some mkfs
1729 * binary needs it. We might want to drop this workaround
1730 * during an unstable branch.
1732 filp
->f_flags
|= O_LARGEFILE
;
1734 if (filp
->f_flags
& O_NDELAY
)
1735 filp
->f_mode
|= FMODE_NDELAY
;
1736 if (filp
->f_flags
& O_EXCL
)
1737 filp
->f_mode
|= FMODE_EXCL
;
1738 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1739 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1741 bdev
= bd_acquire(inode
);
1745 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1747 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1750 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1752 struct gendisk
*disk
= bdev
->bd_disk
;
1753 struct block_device
*victim
= NULL
;
1755 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1757 bdev
->bd_part_count
--;
1759 if (!--bdev
->bd_openers
) {
1760 WARN_ON_ONCE(bdev
->bd_holders
);
1761 sync_blockdev(bdev
);
1764 bdev_write_inode(bdev
);
1766 if (bdev
->bd_contains
== bdev
) {
1767 if (disk
->fops
->release
)
1768 disk
->fops
->release(disk
, mode
);
1770 if (!bdev
->bd_openers
) {
1771 struct module
*owner
= disk
->fops
->owner
;
1773 disk_put_part(bdev
->bd_part
);
1774 bdev
->bd_part
= NULL
;
1775 bdev
->bd_disk
= NULL
;
1776 if (bdev
!= bdev
->bd_contains
)
1777 victim
= bdev
->bd_contains
;
1778 bdev
->bd_contains
= NULL
;
1783 mutex_unlock(&bdev
->bd_mutex
);
1786 __blkdev_put(victim
, mode
, 1);
1789 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1791 mutex_lock(&bdev
->bd_mutex
);
1793 if (mode
& FMODE_EXCL
) {
1797 * Release a claim on the device. The holder fields
1798 * are protected with bdev_lock. bd_mutex is to
1799 * synchronize disk_holder unlinking.
1801 spin_lock(&bdev_lock
);
1803 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1804 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1806 /* bd_contains might point to self, check in a separate step */
1807 if ((bdev_free
= !bdev
->bd_holders
))
1808 bdev
->bd_holder
= NULL
;
1809 if (!bdev
->bd_contains
->bd_holders
)
1810 bdev
->bd_contains
->bd_holder
= NULL
;
1812 spin_unlock(&bdev_lock
);
1815 * If this was the last claim, remove holder link and
1816 * unblock evpoll if it was a write holder.
1818 if (bdev_free
&& bdev
->bd_write_holder
) {
1819 disk_unblock_events(bdev
->bd_disk
);
1820 bdev
->bd_write_holder
= false;
1825 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1826 * event. This is to ensure detection of media removal commanded
1827 * from userland - e.g. eject(1).
1829 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1831 mutex_unlock(&bdev
->bd_mutex
);
1833 __blkdev_put(bdev
, mode
, 0);
1835 EXPORT_SYMBOL(blkdev_put
);
1837 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1839 struct block_device
*bdev
= I_BDEV(bdev_file_inode(filp
));
1840 blkdev_put(bdev
, filp
->f_mode
);
1844 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1846 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1847 fmode_t mode
= file
->f_mode
;
1850 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1851 * to updated it before every ioctl.
1853 if (file
->f_flags
& O_NDELAY
)
1854 mode
|= FMODE_NDELAY
;
1856 mode
&= ~FMODE_NDELAY
;
1858 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1862 * Write data to the block device. Only intended for the block device itself
1863 * and the raw driver which basically is a fake block device.
1865 * Does not take i_mutex for the write and thus is not for general purpose
1868 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1870 struct file
*file
= iocb
->ki_filp
;
1871 struct inode
*bd_inode
= bdev_file_inode(file
);
1872 loff_t size
= i_size_read(bd_inode
);
1873 struct blk_plug plug
;
1876 if (bdev_read_only(I_BDEV(bd_inode
)))
1879 if (!iov_iter_count(from
))
1882 if (iocb
->ki_pos
>= size
)
1885 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1887 blk_start_plug(&plug
);
1888 ret
= __generic_file_write_iter(iocb
, from
);
1890 ret
= generic_write_sync(iocb
, ret
);
1891 blk_finish_plug(&plug
);
1894 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1896 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1898 struct file
*file
= iocb
->ki_filp
;
1899 struct inode
*bd_inode
= bdev_file_inode(file
);
1900 loff_t size
= i_size_read(bd_inode
);
1901 loff_t pos
= iocb
->ki_pos
;
1907 iov_iter_truncate(to
, size
);
1908 return generic_file_read_iter(iocb
, to
);
1910 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1913 * Try to release a page associated with block device when the system
1914 * is under memory pressure.
1916 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1918 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1920 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1921 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1923 return try_to_free_buffers(page
);
1926 static int blkdev_writepages(struct address_space
*mapping
,
1927 struct writeback_control
*wbc
)
1929 if (dax_mapping(mapping
)) {
1930 struct block_device
*bdev
= I_BDEV(mapping
->host
);
1932 return dax_writeback_mapping_range(mapping
, bdev
, wbc
);
1934 return generic_writepages(mapping
, wbc
);
1937 static const struct address_space_operations def_blk_aops
= {
1938 .readpage
= blkdev_readpage
,
1939 .readpages
= blkdev_readpages
,
1940 .writepage
= blkdev_writepage
,
1941 .write_begin
= blkdev_write_begin
,
1942 .write_end
= blkdev_write_end
,
1943 .writepages
= blkdev_writepages
,
1944 .releasepage
= blkdev_releasepage
,
1945 .direct_IO
= blkdev_direct_IO
,
1946 .is_dirty_writeback
= buffer_check_dirty_writeback
,
1949 #define BLKDEV_FALLOC_FL_SUPPORTED \
1950 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1951 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1953 static long blkdev_fallocate(struct file
*file
, int mode
, loff_t start
,
1956 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1957 struct address_space
*mapping
;
1958 loff_t end
= start
+ len
- 1;
1962 /* Fail if we don't recognize the flags. */
1963 if (mode
& ~BLKDEV_FALLOC_FL_SUPPORTED
)
1966 /* Don't go off the end of the device. */
1967 isize
= i_size_read(bdev
->bd_inode
);
1971 if (mode
& FALLOC_FL_KEEP_SIZE
) {
1972 len
= isize
- start
;
1973 end
= start
+ len
- 1;
1979 * Don't allow IO that isn't aligned to logical block size.
1981 if ((start
| len
) & (bdev_logical_block_size(bdev
) - 1))
1984 /* Invalidate the page cache, including dirty pages. */
1985 mapping
= bdev
->bd_inode
->i_mapping
;
1986 truncate_inode_pages_range(mapping
, start
, end
);
1989 case FALLOC_FL_ZERO_RANGE
:
1990 case FALLOC_FL_ZERO_RANGE
| FALLOC_FL_KEEP_SIZE
:
1991 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
1992 GFP_KERNEL
, BLKDEV_ZERO_NOUNMAP
);
1994 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
:
1995 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
1996 GFP_KERNEL
, BLKDEV_ZERO_NOFALLBACK
);
1998 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
| FALLOC_FL_NO_HIDE_STALE
:
1999 error
= blkdev_issue_discard(bdev
, start
>> 9, len
>> 9,
2009 * Invalidate again; if someone wandered in and dirtied a page,
2010 * the caller will be given -EBUSY. The third argument is
2011 * inclusive, so the rounding here is safe.
2013 return invalidate_inode_pages2_range(mapping
,
2014 start
>> PAGE_SHIFT
,
2018 const struct file_operations def_blk_fops
= {
2019 .open
= blkdev_open
,
2020 .release
= blkdev_close
,
2021 .llseek
= block_llseek
,
2022 .read_iter
= blkdev_read_iter
,
2023 .write_iter
= blkdev_write_iter
,
2024 .mmap
= generic_file_mmap
,
2025 .fsync
= blkdev_fsync
,
2026 .unlocked_ioctl
= block_ioctl
,
2027 #ifdef CONFIG_COMPAT
2028 .compat_ioctl
= compat_blkdev_ioctl
,
2030 .splice_read
= generic_file_splice_read
,
2031 .splice_write
= iter_file_splice_write
,
2032 .fallocate
= blkdev_fallocate
,
2035 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
2038 mm_segment_t old_fs
= get_fs();
2040 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
2045 EXPORT_SYMBOL(ioctl_by_bdev
);
2048 * lookup_bdev - lookup a struct block_device by name
2049 * @pathname: special file representing the block device
2051 * Get a reference to the blockdevice at @pathname in the current
2052 * namespace if possible and return it. Return ERR_PTR(error)
2055 struct block_device
*lookup_bdev(const char *pathname
)
2057 struct block_device
*bdev
;
2058 struct inode
*inode
;
2062 if (!pathname
|| !*pathname
)
2063 return ERR_PTR(-EINVAL
);
2065 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
2067 return ERR_PTR(error
);
2069 inode
= d_backing_inode(path
.dentry
);
2071 if (!S_ISBLK(inode
->i_mode
))
2074 if (!may_open_dev(&path
))
2077 bdev
= bd_acquire(inode
);
2084 bdev
= ERR_PTR(error
);
2087 EXPORT_SYMBOL(lookup_bdev
);
2089 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
2091 struct super_block
*sb
= get_super(bdev
);
2096 * no need to lock the super, get_super holds the
2097 * read mutex so the filesystem cannot go away
2098 * under us (->put_super runs with the write lock
2101 shrink_dcache_sb(sb
);
2102 res
= invalidate_inodes(sb
, kill_dirty
);
2105 invalidate_bdev(bdev
);
2108 EXPORT_SYMBOL(__invalidate_device
);
2110 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
2112 struct inode
*inode
, *old_inode
= NULL
;
2114 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2115 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
2116 struct address_space
*mapping
= inode
->i_mapping
;
2117 struct block_device
*bdev
;
2119 spin_lock(&inode
->i_lock
);
2120 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
2121 mapping
->nrpages
== 0) {
2122 spin_unlock(&inode
->i_lock
);
2126 spin_unlock(&inode
->i_lock
);
2127 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
2129 * We hold a reference to 'inode' so it couldn't have been
2130 * removed from s_inodes list while we dropped the
2131 * s_inode_list_lock We cannot iput the inode now as we can
2132 * be holding the last reference and we cannot iput it under
2133 * s_inode_list_lock. So we keep the reference and iput it
2138 bdev
= I_BDEV(inode
);
2140 mutex_lock(&bdev
->bd_mutex
);
2141 if (bdev
->bd_openers
)
2143 mutex_unlock(&bdev
->bd_mutex
);
2145 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2147 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);